P2.64
SNOW-LIQUID RATIOS OVER THE NORTHERN SIERRA NEVADA: CLIMATOLOGY
AND A PREDICTIVE METHODOLOGY
Bill M. Rasch1 *, Darren T. Van Cleave1,2
1. National Weather Service Sacramento, CA
2. National Weather Service Missoula, MT
1.
INTRODUCTION
Snowstorms over the Northern Sierra can have
significant impacts on the local and regional economy
due to effects of slow-downs or closures of Interstate 80
which functions as a major artery of interstate
commerce. According to CalTrans (California
Transportation Department), a one hour shutdown of I80 over Donner Pass can result in losses exceeding 1
million dollars (Eric Kurth, personal communication).
Lane closures and shutdowns are generally a function
of both snowfall and snow water content. Thus,
forecasts of snowfall and snow characteristics are of
great importance.
Snow forecasting in areas of complex terrain and
varying climatic zones generally consists of three
components; quantitative precipitation forecast (QPF),
snow-to-liquid ratio (SLR), and snow level. When
National Weather Service (NWS) forecasters created
gridded snowfall forecasts, they are usually also making
forecasts of these three elements. Historically, the NWS
office in Sacramento placed much attention on QPF and
snow level forecasts, while the SLR was either created
via rule-of-thumb, forecaster best guess, or local tools
relating SLR to elevation. In fact, the SLR was
sometimes increased to physically unrealistic values in
order to compensate for QPF guidance being too low.
Unfortunately, these methodologies are not scientifically
sound and in practice produce mixed results. A more
scientifically sound method to create SLR would ideally
involve some type of cloud physics responsible for ice
crystal growth, as well as recognition of crystal
modification if melting is involved. This research will
attempt to create a tool using such a method.
Previous studies have attempted to link SLR to
atmospheric variables, either from radiosonde data or
forecast models (Diamond and Lowry, 1956; Alcott and
Steenburgh, 2010). Diamond and Lowry researched
* Corresponding author address: Bill M. Rasch,
National Weather Service, 3310 El Camino Ave,
Sacramento, CA 95821; e-mail:
William.Rasch@noaa.gov
SLR at the Sierra Snow Lab near Donner Pass, CA (a
location near the two sites used in this study), and found
a modest correlation (R =0.64) between 700mb
temperature from upper-air soundings in Oakland, CA
and snow-density (a relative of SLR). Many years later,
Alcott and Steenburgh were able to correlate 650mb
temperature (R=-0.62) and winds (R=-0.39) from model
data over the Wasatch mountains of Utah to SLR at a
ski area near Salt Lake City.
The underlying idea behind connecting temperatures
aloft to SLR lies in the microphysical properties of ice
crystal formation within clouds. Libbrecht (1999)
demonstrated the various types of ice crystal
development as a function of temperature. Most notably,
dendrite growth (which supports higher SLR’s due to
lower density) is preferred at temperatures of -12 to 18c, with a gradual shift from lower-SLR columns and
needles to higher-SLR dendrites as temperature
decreases (assuming adequate supersaturation). SLR
then tends to decrease at colder temperatures as crystal
tendency trends back towards columns. However,
temperatures found in snow-producing clouds over the
Sierra do not frequently reach these colder columnfavoring temperatures due to the influence of warmer
maritime airmasses.
2. DATA
Measurements of snow and melted liquid
precipitation were retrieved from the CoCoRaHS
network (Cifelli et al, 2005) at Kingvale and Soda
Springs, as marked in Fig. 1. One of the data sources,
Kingvale, happened to also be a reliable weather spotter
for NWS Sacramento, which provided extra confidence
in the data. The Sierra Snow Laboratory, located
nearby, was also briefly considered as a data source
due to its long history of snow observing as well as its
use in a related study (Diamond & Lowry, 1956). It was
ultimately excluded, however, due to differences in
measurement and recording methods. It is included in
the figure for reference. Lastly, data from Blue Canyon
were used as a quality control reference for the
CoCoRaHS locations.
Figure 1 - Location of observations along Interstate 80.
Cases of snowfall greater than 2 inches were
retrieved for four winter seasons dating back to 2008 for
Kingvale and Soda Springs. Cases with mixed rain and
snow were removed to eliminate contamination of the
dataset. The existence of mixed precipitation was
inferred from temperature and weather observations at
the nearby Blue Canyon ASOS and weather
observations from the Sierra Snow Lab. Blue Canyon
lies around a thousand feet lower than CoCoRaHS
stations, while the Sierra Snow Lab is around 500 feet
higher than the stations. For each case, if the Sierra
Snow Lab reported any liquid, the case was thrown out.
If Blue Canyon reported all snow, the case was
included. If Blue Canyon reported mixed rain and snow
while the Sierra Snow Lab reported all snow, the
surface temperature at Blue Canyon was interpolated
along a moist adiabat to Kingvale’s elevation. If this
process yielded below-freezing temperatures, the case
was included.
Official NWS Sacramento gridded forecasts of snow
and QPF were retrieved from local office records
through the BOIVerify program commonly used at NWS
offices. Unfortunately, this dataset was temporally
limited due to disk space constraints, only going as far
back as winter 2011. Thus, the full CoCoRaHS dataset
was used for examining the SLR climatology, while only
the combined period of record dating back to February
2011 was used for the SLR forecasting methodology
and verification.
3. SNOW RATIO CLIMATOLOGY
Histograms of the overall snow ratios for both sites
were created, which showed a median SLR value of 9
with relatively few cases reaching above 20. This is
remarkably similar to Baxter et. al (2005) who found an
average of 9 over the Sacramento forecast area despite
using a much different dataset over a different period of
time (30 years of data from NWS cooperative observer
data). Histograms by month for one of the stations,
Kingvale, are provided in Fig. 2. As would be expected,
SLR trends upward towards the colder months in the
middle of winter and visa-versa in the spring. While
February shows the highest average SLR, December is
the month with the most extreme cases of SLR greater
than 20.
Figure 2 - Kingvale SLR by month.
4. TOOL METHODOLOGY
Following Alcott and Steenburgh (2010) and
Diamond and Lowry (1954), four winter seasons of
CoCoRaHS SLR data were correlated to modeled winds
and temperatures aloft. The North American Regional
Reanalysis (NARR; Mesinger et al. 2006) model was
selected due to its temporal coverage and ease of use.
Figure 3 - 700mb temperature versus SLR at Soda Springs.
A grid point between Kingvale and Soda Springs was
selected to be the representative model point. Winds
and temperatures from 800-600mb were tested for
correlation with SLR at both Kingvale and Soda Springs.
Before correlations were computed, the CoCoRaHS
data was filtered by removing SLRs more than four
standard deviations from the mean (this removed the
few strongest outliers which may have been bad data
points). The best correlations for both wind magnitude
and temperature were found at 700mb; this happens to
be the same geopotential height used by Diamond and
Lowry (1954). Fig. 3 shows the 700mb correlation to
SLR at Soda Springs, which exhibits a nearly uniform
spread across all 700mb temperatures.
While this 700mb relationship works reasonably well
for ice crystals unaltered during their decent to the
surface, modification by elevated warm layers,
refreezing in surface cold pools, and many other
interactions can introduce additional challenges. For the
western slope of the Northern Sierra Nevada, melting of
snow at lower elevations and eventual transition to rain
is a common feature of virtually all winter storms in the
area. Thus, an SLR methodology should include some
acknowledgement of the contribution of surface
conditions. Local NWS Sacramento forecasting
knowledge as well as previous research shows that SLR
has some relationship with surface temperature (Judson
and Doesken, 2000). For this study, hourly temperature
forecasts were compared to SLR data, with the hourly
temperatures averaged over 12 hour periods. A modest
correlation of R2=0.41 at Kingvale and 0.27 at Soda
Springs was found. This result is within range of the
findings of Judson & Doesken (R2=0.27, 2000). While
the correlation is not as strong and consistent as the
700mb relationship, it was considered robust enough to
be used as a second component to an eventual SLR
forecasting tool.
Figure 4 - Conceptual schematic of SLR tool
A conceptual model for a SLR forecasting tool is
depicted in Fig. 4. The 700mb relationship is used at
elevations above the snow level (and for below-freezing
temperatures in the hourly weather grids) where cloud
physics are the driving factor in SLR; below this level,
the surface relationship is used. By connecting SLR to
surface temperature near the freezing level, interelement consistency between the weather, hourly
temperature, snow amount, and snow level grids is
preserved (assuming the snow level grid is used to
create the rain/snow transition zone in the
Figure 5 – Magnitude of errors in snow forecasts at Kingvale from 11/2011 to 3/2013. Each point is an event with > 2”.
weather grid).
A tool was created at NWS Sacramento within the
Graphical Forecast Editor (GFE) to perform the
previously mentioned methodology. With default
settings, this tool calculates SLR at all locations using
the 700mb relationship, calculates SLR using the
surface temperature relationship where the average
hourly temperature is forecasted to be below 35, divides
SLR in half where the weather grid contains mixed rain
and snow, and removes SLR where the weather grid
contains no snow. Several options are provided to the
forecaster, including the selection of models for the
700mb portion and the ability to select the 700mb or
surface relationships only. This helps to encompass the
wide range of operational situations which fall outside
the bounds assumed in the aforementioned conceptual
model.
5. RESULTS
Using the methodology described above, output from
the SLR tool was compared to official NWS forecasts.
The forecast errors (|observed-forecasted|) for both the
NWS forecast and the SLR tool forecast at Kingvale are
plotted in Figure 5. While the SLR tool does not always
perform better than the official forecast (actually
performing worse on occasion), it does, on average,
reduce the magnitude of errors, particularly the ‘big
busts’ (errors > 15 inches). Considering both Kingvale
and Soda Springs together for winter 2011, the SLR tool
offered an average 22% reduction in forecast error.
More notably, for cases over 10 inches, the tool yielded
average improvements of 32%.
Results of the GFE tool are shown in Fig. 6 for a
snow event in December 2012. The official forecast
using the new tool resulted in a far better snow forecast,
which also happened to better coordinate with a
neighboring NWS forecast office.
6. FUTURE IMPROVEMENTS
While this GFE tool represents an improvement in
the scientific integrity of SLR forecasts and resulting
snow forecasts, it has multiple areas for improvement.
Practical limitations in this research project prevented
the direct use of NAM, GFS, and ECMWF forecasts;
namely, those models were not easily available for quick
download via scripts as was the NARR. The GFE tool
created through this research thus makes the inherent
assumption that correlations of NARR data to SLR are
applicable for other models; this may not necessarily be
the case. Ideally, each model would need its own set of
correlations. Fortunately, the NAM model is fairly similar
to the NARR as the NARR’s physics consist of the NAM
as used in the early 2000’s (Messinger, 2006).
Another weakness in forming this tool is using one
climatic zone to forecast SLR for an entire forecast area.
Unfortunately, there are not many sites in the Sierra
which have reliable snowfall and melted snow data on a
routine basis, particularly at the higher elevations such
as Donner Pass. Ideally, data would be retrieved for
several points representing the various climatic zones of
the forecast area, and correlations could then be
included in the tool which would vary spatially and allow
for appropriate climatological variability within the
forecast domain. With plentiful data, correlations could
even be performed for different seasons to add another
layer of precision.
Figure 6 - Upper panels: SLR and snow forecasts within GFE using existing methodologies for a December storm. Lower panel:
same forecasts, except using SLRs created with the new tool.
One particular difficulty encountered in forming the
tool was handling the transition from the 700mb
relationship to the surface relationship. Ideally, the tool
would transition from one method to the other via a
small transition zone where the rate of change of the
values is tied to the topography. This was beyond the
skills of those working on this project, so instead the tool
creates masks for each portions, then simply overlays
them. This can create a rapid transition in SLR from
higher to lower values as elevation decreases, which
may not be reflective of nature.
After testing the 700mb relationship, surface
relationship, and various models and blends within the
tool over early winter 2012/2013, it quickly became
apparent that a one-size-fits-all approach to the tool
options was simply impossible. After testing, the optimal
approach was to allow some options to the forecaster to
tailor the tool to the situation (as opposed to a “black
box” approach where the forecaster simply runs a tool
with default selections built in and no options allowed).
ACKNOWLEDGEMENTS
Local undergraduate volunteers at NWS Sacramento
Brian Rico and Matt Little collected data for the study.
Also, NWS Sacramento forecaster Eric Kurth retrieved
data from CalTrans and the Sierra Snow Lab.
Cifelli, R., N. Doesken, P. Kennedy, L. D. Carey, S.
Rutledge, C. Gimmestad, and T. Depue. 2005. The
community collaborative rain, hail, and snow network.
Bull. Amer. Meteor. Soc, 86:1069–1077.
Diamond, M., and W. P. Lowry, 1954: Correlation of
density of new snow with 700-millibar temperature. J.
Meteor., 11, 512–513.
REFERENCES
Judson, A., and N. Doesken, 2000: Density of freshly
fallen snow in the central Rocky Mountains. Bull. Amer.
Meteor. Soc., 81, 1577–1587.
Alcott, T. I., and W. J. Steenburgh, 2010: Snow-to-liquid
variability and prediction at a high-elevation site in
Utah’s Wasatch Mountains. Wea. Forecasting, 25, 323–
337.
Libbrecht, K. G., 1999: A Guide to Snowflakes.
[Available on line at:
http://www.its.caltech.edu/~atomic/snowcrystals/].
Baxter, A. M., C. E. Graves, and J. T. Moore, 2005: A
climatology of snow-to-liquid ratio for the contiguous
United States. Wea. Forecasting, 20, 729–744.
Mesinger, F., and Coauthors, 2006: North American
regional reanalysis. Bull. Amer. Meteor. Soc., 87, 343–
360.